Sign language gesture recognition with bispectrum features using SVM

. Wi-Fi based sensing system captures the signal reflections due to human gestures as Channel State Information (CSI) values in subcarrier level for accurately predicting the fine-grained gestures. The proposed work explores the Higher Order Statistical (HOS) method by deriving bispectrum features (BF) from raw signal by adopting a Conditional Informative Feature Extraction (CIFE) technique from information theory to form a subset of informative and best features. Support Vector Machine (SVM) classifier is adopted in the present work for classifying the gesture and to measure the prediction accuracy. The present work is validated on a secondary dataset, SignFi, having data collected from two different environments with varying number of users and sign gestures. SVM reports an overall accuracy of 83.8%, 94.1%, 74.9% and 75.6% in different environments/scenarios.